Fusjon av informasjon fra perimeterovervåkingssensorer

Abstract

This report first describes in detail three standard methods for sensor fusion; rule based fusion, Bayesian fusion, and Dempster Shafer fusion. Next, it is described how information from various surveillance sensors can be utilized by applying these techniques. Available sensors were microphonic cable, geophones, passive infrared detector, and an electromagnetic sensor (Repels). Several experiments have been carried out. First we present results from experiments where four persons have made intrusions in various ways; “careful walking”, high and low crawling, rolling. Next, processing results of two background recordings (both made during daytime) are shown. Finally, a long term test are presented. The experiments show an increased performance by using fused information from several sensors (compared to single sensor performance); the detection rate increases, and at the same time the false detection rate does not show any increase (if we demand that one sensor alone cannot give any detection, we will still have the same high detection rate, but the false detection rate is drastically reduced). This leads to the conclusion that a perimeter surveillance system will increase its performance by using more than one sensor together with the sensor fusion techniques described in this report.